"""
A collection of utility functions and classes. Originally, many
(but not all) were from the Python Cookbook -- hence the name cbook.
"""
import collections
import collections.abc
import contextlib
import functools
import gzip
import itertools
import math
import operator
import os
from pathlib import Path
import shlex
import subprocess
import sys
import time
import traceback
import types
import weakref
import numpy as np
try:
from numpy.exceptions import VisibleDeprecationWarning # numpy >= 1.25
except ImportError:
from numpy import VisibleDeprecationWarning
import matplotlib
from matplotlib import _api, _c_internal_utils, mlab
class _ExceptionInfo:
"""
A class to carry exception information around.
This is used to store and later raise exceptions. It's an alternative to
directly storing Exception instances that circumvents traceback-related
issues: caching tracebacks can keep user's objects in local namespaces
alive indefinitely, which can lead to very surprising memory issues for
users and result in incorrect tracebacks.
"""
def __init__(self, cls, *args, notes=None):
self._cls = cls
self._args = args
self._notes = notes if notes is not None else []
@classmethod
def from_exception(cls, exc):
return cls(type(exc), *exc.args, notes=getattr(exc, "__notes__", []))
def to_exception(self):
exc = self._cls(*self._args)
for note in self._notes:
exc.add_note(note)
return exc
def _get_running_interactive_framework():
"""
Return the interactive framework whose event loop is currently running, if
any, or "headless" if no event loop can be started, or None.
Returns
-------
Optional[str]
One of the following values: "qt", "gtk3", "gtk4", "wx", "tk",
"macosx", "headless", ``None``.
"""
# Use ``sys.modules.get(name)`` rather than ``name in sys.modules`` as
# entries can also have been explicitly set to None.
QtWidgets = (
sys.modules.get("PyQt6.QtWidgets")
or sys.modules.get("PySide6.QtWidgets")
or sys.modules.get("PyQt5.QtWidgets")
or sys.modules.get("PySide2.QtWidgets")
)
if QtWidgets and QtWidgets.QApplication.instance():
return "qt"
Gtk = sys.modules.get("gi.repository.Gtk")
if Gtk:
if Gtk.MAJOR_VERSION == 4:
from gi.repository import GLib
if GLib.main_depth():
return "gtk4"
if Gtk.MAJOR_VERSION == 3 and Gtk.main_level():
return "gtk3"
wx = sys.modules.get("wx")
if wx and wx.GetApp():
return "wx"
tkinter = sys.modules.get("tkinter")
if tkinter:
codes = {tkinter.mainloop.__code__, tkinter.Misc.mainloop.__code__}
for frame in sys._current_frames().values():
while frame:
if frame.f_code in codes:
return "tk"
frame = frame.f_back
# Preemptively break reference cycle between locals and the frame.
del frame
macosx = sys.modules.get("matplotlib.backends._macosx")
if macosx and macosx.event_loop_is_running():
return "macosx"
if not _c_internal_utils.display_is_valid():
return "headless"
return None
def _exception_printer(exc):
if _get_running_interactive_framework() in ["headless", None]:
raise exc
else:
traceback.print_exc()
class _StrongRef:
"""
Wrapper similar to a weakref, but keeping a strong reference to the object.
"""
def __init__(self, obj):
self._obj = obj
def __call__(self):
return self._obj
def __eq__(self, other):
return isinstance(other, _StrongRef) and self._obj == other._obj
def __hash__(self):
return hash(self._obj)
def _weak_or_strong_ref(func, callback):
"""
Return a `WeakMethod` wrapping *func* if possible, else a `_StrongRef`.
"""
try:
return weakref.WeakMethod(func, callback)
except TypeError:
return _StrongRef(func)
class _UnhashDict:
"""
A minimal dict-like class that also supports unhashable keys, storing them
in a list of key-value pairs.
This class only implements the interface needed for `CallbackRegistry`, and
tries to minimize the overhead for the hashable case.
"""
def __init__(self, pairs):
self._dict = {}
self._pairs = []
for k, v in pairs:
self[k] = v
def __setitem__(self, key, value):
try:
self._dict[key] = value
except TypeError:
for i, (k, v) in enumerate(self._pairs):
if k == key:
self._pairs[i] = (key, value)
break
else:
self._pairs.append((key, value))
def __getitem__(self, key):
try:
return self._dict[key]
except TypeError:
pass
for k, v in self._pairs:
if k == key:
return v
raise KeyError(key)
def pop(self, key, *args):
try:
if key in self._dict:
return self._dict.pop(key)
except TypeError:
for i, (k, v) in enumerate(self._pairs):
if k == key:
del self._pairs[i]
return v
if args:
return args[0]
raise KeyError(key)
def __iter__(self):
yield from self._dict
for k, v in self._pairs:
yield k
class CallbackRegistry:
"""
Handle registering, processing, blocking, and disconnecting
for a set of signals and callbacks:
>>> def oneat(x):
... print('eat', x)
>>> def ondrink(x):
... print('drink', x)
>>> from matplotlib.cbook import CallbackRegistry
>>> callbacks = CallbackRegistry()
>>> id_eat = callbacks.connect('eat', oneat)
>>> id_drink = callbacks.connect('drink', ondrink)
>>> callbacks.process('drink', 123)
drink 123
>>> callbacks.process('eat', 456)
eat 456
>>> callbacks.process('be merry', 456) # nothing will be called
>>> callbacks.disconnect(id_eat)
>>> callbacks.process('eat', 456) # nothing will be called
>>> with callbacks.blocked(signal='drink'):
... callbacks.process('drink', 123) # nothing will be called
>>> callbacks.process('drink', 123)
drink 123
>>> callbacks.disconnect(ondrink, signal='drink') # disconnect by func
>>> callbacks.process('drink', 123) # nothing will be called
In practice, one should always disconnect all callbacks when they are
no longer needed to avoid dangling references (and thus memory leaks).
However, real code in Matplotlib rarely does so, and due to its design,
it is rather difficult to place this kind of code. To get around this,
and prevent this class of memory leaks, we instead store weak references
to bound methods only, so when the destination object needs to die, the
CallbackRegistry won't keep it alive.
Parameters
----------
exception_handler : callable, optional
If not None, *exception_handler* must be a function that takes an
`Exception` as single parameter. It gets called with any `Exception`
raised by the callbacks during `CallbackRegistry.process`, and may
either re-raise the exception or handle it in another manner.
The default handler prints the exception (with `traceback.print_exc`) if
an interactive event loop is running; it re-raises the exception if no
interactive event loop is running.
signals : list, optional
If not None, *signals* is a list of signals that this registry handles:
attempting to `process` or to `connect` to a signal not in the list
throws a `ValueError`. The default, None, does not restrict the
handled signals.
"""
# We maintain two mappings:
# callbacks: signal -> {cid -> weakref-to-callback}
# _func_cid_map: {(signal, weakref-to-callback) -> cid}
def __init__(self, exception_handler=_exception_printer, *, signals=None):
self._signals = None if signals is None else list(signals) # Copy it.
self.exception_handler = exception_handler
self.callbacks = {}
self._cid_gen = itertools.count()
self._func_cid_map = _UnhashDict([])
# A hidden variable that marks cids that need to be pickled.
self._pickled_cids = set()
def __getstate__(self):
return {
**vars(self),
# In general, callbacks may not be pickled, so we just drop them,
# unless directed otherwise by self._pickled_cids.
"callbacks": {s: {cid: proxy() for cid, proxy in d.items()
if cid in self._pickled_cids}
for s, d in self.callbacks.items()},
# It is simpler to reconstruct this from callbacks in __setstate__.
"_func_cid_map": None,
"_cid_gen": next(self._cid_gen)
}
def __setstate__(self, state):
cid_count = state.pop('_cid_gen')
vars(self).update(state)
self.callbacks = {
s: {cid: _weak_or_strong_ref(func, functools.partial(self._remove_proxy, s))
for cid, func in d.items()}
for s, d in self.callbacks.items()}
self._func_cid_map = _UnhashDict(
((s, proxy), cid)
for s, d in self.callbacks.items() for cid, proxy in d.items())
self._cid_gen = itertools.count(cid_count)
def connect(self, signal, func):
"""Register *func* to be called when signal *signal* is generated."""
if self._signals is not None:
_api.check_in_list(self._signals, signal=signal)
proxy = _weak_or_strong_ref(func, functools.partial(self._remove_proxy, signal))
try:
return self._func_cid_map[signal, proxy]
except KeyError:
cid = self._func_cid_map[signal, proxy] = next(self._cid_gen)
self.callbacks.setdefault(signal, {})[cid] = proxy
return cid
def _connect_picklable(self, signal, func):
"""
Like `.connect`, but the callback is kept when pickling/unpickling.
Currently internal-use only.
"""
cid = self.connect(signal, func)
self._pickled_cids.add(cid)
return cid
# Keep a reference to sys.is_finalizing, as sys may have been cleared out
# at that point.
def _remove_proxy(self, signal, proxy, *, _is_finalizing=sys.is_finalizing):
if _is_finalizing():
# Weakrefs can't be properly torn down at that point anymore.
return
cid = self._func_cid_map.pop((signal, proxy), None)
if cid is not None:
del self.callbacks[signal][cid]
self._pickled_cids.discard(cid)
else: # Not found
return
if len(self.callbacks[signal]) == 0: # Clean up empty dicts
del self.callbacks[signal]
@_api.rename_parameter("3.11", "cid", "cid_or_func")
def disconnect(self, cid_or_func, *, signal=None):
"""
Disconnect a callback.
Parameters
----------
cid_or_func : int or callable
If an int, disconnect the callback with that connection id.
If a callable, disconnect that function from signals.
signal : optional
Only used when *cid_or_func* is a callable. If given, disconnect
the function only from that specific signal. If not given,
disconnect from all signals the function is connected to.
Notes
-----
No error is raised if such a callback does not exist.
"""
if isinstance(cid_or_func, int):
if signal is not None:
raise ValueError(
"signal cannot be specified when disconnecting by cid")
for sig, proxy in self._func_cid_map:
if self._func_cid_map[sig, proxy] == cid_or_func:
break
else: # Not found
return
self._remove_proxy(sig, proxy)
elif signal is not None:
# Disconnect from a specific signal
proxy = _weak_or_strong_ref(cid_or_func, None)
self._remove_proxy(signal, proxy)
else:
# Disconnect from all signals
proxy = _weak_or_strong_ref(cid_or_func, None)
for sig, prx in list(self._func_cid_map):
if prx == proxy:
self._remove_proxy(sig, proxy)
def process(self, s, *args, **kwargs):
"""
Process signal *s*.
All of the functions registered to receive callbacks on *s* will be
called with ``*args`` and ``**kwargs``.
"""
if self._signals is not None:
_api.check_in_list(self._signals, signal=s)
for ref in list(self.callbacks.get(s, {}).values()):
func = ref()
if func is not None:
try:
func(*args, **kwargs)
# this does not capture KeyboardInterrupt, SystemExit,
# and GeneratorExit
except Exception as exc:
if self.exception_handler is not None:
self.exception_handler(exc)
else:
raise
@contextlib.contextmanager
def blocked(self, *, signal=None):
"""
Block callback signals from being processed.
A context manager to temporarily block/disable callback signals
from being processed by the registered listeners.
Parameters
----------
signal : str, optional
The callback signal to block. The default is to block all signals.
"""
orig = self.callbacks
try:
if signal is None:
# Empty out the callbacks
self.callbacks = {}
else:
# Only remove the specific signal
self.callbacks = {k: orig[k] for k in orig if k != signal}
yield
finally:
self.callbacks = orig
class silent_list(list):
"""
A list with a short ``repr()``.
This is meant to be used for a homogeneous list of artists, so that they
don't cause long, meaningless output.
Instead of ::
[,
,
]
one will get ::
If ``self.type`` is None, the type name is obtained from the first item in
the list (if any).
"""
def __init__(self, type, seq=None):
self.type = type
if seq is not None:
self.extend(seq)
def __repr__(self):
if self.type is not None or len(self) != 0:
tp = self.type if self.type is not None else type(self[0]).__name__
return f""
else:
return ""
def _local_over_kwdict(
local_var, kwargs, *keys,
warning_cls=_api.MatplotlibDeprecationWarning):
out = local_var
for key in keys:
kwarg_val = kwargs.pop(key, None)
if kwarg_val is not None:
if out is None:
out = kwarg_val
else:
_api.warn_external(f'"{key}" keyword argument will be ignored',
warning_cls)
return out
def strip_math(s):
"""
Remove latex formatting from mathtext.
Only handles fully math and fully non-math strings.
"""
if len(s) >= 2 and s[0] == s[-1] == "$":
s = s[1:-1]
for tex, plain in [
(r"\times", "x"), # Specifically for Formatter support.
(r"\mathdefault", ""),
(r"\rm", ""),
(r"\cal", ""),
(r"\tt", ""),
(r"\it", ""),
("\\", ""),
("{", ""),
("}", ""),
]:
s = s.replace(tex, plain)
return s
def _strip_comment(s):
"""Strip everything from the first unquoted #."""
pos = 0
while True:
quote_pos = s.find('"', pos)
hash_pos = s.find('#', pos)
if quote_pos < 0:
without_comment = s if hash_pos < 0 else s[:hash_pos]
return without_comment.strip()
elif 0 1:
raise ValueError("Masked arrays must be 1-D")
try:
x = np.asanyarray(x)
except (VisibleDeprecationWarning, ValueError):
# NumPy 1.19 raises a warning about ragged arrays, but we want
# to accept basically anything here.
x = np.asanyarray(x, dtype=object)
if x.ndim == 1:
x = safe_masked_invalid(x)
seqlist[i] = True
if np.ma.is_masked(x):
masks.append(np.ma.getmaskarray(x))
margs.append(x) # Possibly modified.
if len(masks):
mask = np.logical_or.reduce(masks)
for i, x in enumerate(margs):
if seqlist[i]:
margs[i] = np.ma.array(x, mask=mask)
return margs
def _broadcast_with_masks(*args, compress=False):
"""
Broadcast inputs, combining all masked arrays.
Parameters
----------
*args : array-like
The inputs to broadcast.
compress : bool, default: False
Whether to compress the masked arrays. If False, the masked values
are replaced by NaNs.
Returns
-------
list of array-like
The broadcasted and masked inputs.
"""
# extract the masks, if any
masks = [k.mask for k in args if isinstance(k, np.ma.MaskedArray)]
# broadcast to match the shape
bcast = np.broadcast_arrays(*args, *masks)
inputs = bcast[:len(args)]
masks = bcast[len(args):]
if masks:
# combine the masks into one
mask = np.logical_or.reduce(masks)
# put mask on and compress
if compress:
inputs = [np.ma.array(k, mask=mask).compressed()
for k in inputs]
else:
inputs = [np.ma.array(k, mask=mask, dtype=float).filled(np.nan).ravel()
for k in inputs]
else:
inputs = [np.ravel(k) for k in inputs]
return inputs
def boxplot_stats(X, whis=1.5, bootstrap=None, labels=None, autorange=False):
r"""
Return a list of dictionaries of statistics used to draw a series of box
and whisker plots using `~.Axes.bxp`.
Parameters
----------
X : array-like
Data that will be represented in the boxplots. Should have 2 or
fewer dimensions.
whis : float or (float, float), default: 1.5
The position of the whiskers.
If a float, the lower whisker is at the lowest datum above
``Q1 - whis*(Q3-Q1)``, and the upper whisker at the highest datum below
``Q3 + whis*(Q3-Q1)``, where Q1 and Q3 are the first and third
quartiles. The default value of ``whis = 1.5`` corresponds to Tukey's
original definition of boxplots.
If a pair of floats, they indicate the percentiles at which to draw the
whiskers (e.g., (5, 95)). In particular, setting this to (0, 100)
results in whiskers covering the whole range of the data.
In the edge case where ``Q1 == Q3``, *whis* is automatically set to
(0, 100) (cover the whole range of the data) if *autorange* is True.
Beyond the whiskers, data are considered outliers and are plotted as
individual points.
bootstrap : int, optional
Number of times the confidence intervals around the median
should be bootstrapped (percentile method).
labels : list of str, optional
Labels for each dataset. Length must be compatible with
dimensions of *X*.
autorange : bool, optional (False)
When `True` and the data are distributed such that the 25th and 75th
percentiles are equal, ``whis`` is set to (0, 100) such that the
whisker ends are at the minimum and maximum of the data.
Returns
-------
list of dict
A list of dictionaries containing the results for each column
of data. Keys of each dictionary are the following:
======== ===================================
Key Value Description
======== ===================================
label tick label for the boxplot
mean arithmetic mean value
med 50th percentile
q1 first quartile (25th percentile)
q3 third quartile (75th percentile)
iqr interquartile range
cilo lower notch around the median
cihi upper notch around the median
whislo end of the lower whisker
whishi end of the upper whisker
fliers outliers
======== ===================================
Notes
-----
Non-bootstrapping approach to confidence interval uses Gaussian-based
asymptotic approximation:
.. math::
\mathrm{med} \pm 1.57 \times \frac{\mathrm{iqr}}{\sqrt{N}}
General approach from:
McGill, R., Tukey, J.W., and Larsen, W.A. (1978) "Variations of
Boxplots", The American Statistician, 32:12-16.
"""
def _bootstrap_median(data, N=5000):
# determine 95% confidence intervals of the median
M = len(data)
percentiles = [2.5, 97.5]
bs_index = np.random.randint(M, size=(N, M))
bsData = data[bs_index]
estimate = np.median(bsData, axis=1, overwrite_input=True)
CI = np.percentile(estimate, percentiles)
return CI
def _compute_conf_interval(data, med, iqr, bootstrap):
if bootstrap is not None:
# Do a bootstrap estimate of notch locations.
# get conf. intervals around median
CI = _bootstrap_median(data, N=bootstrap)
notch_min = CI[0]
notch_max = CI[1]
else:
N = len(data)
notch_min = med - 1.57 * iqr / np.sqrt(N)
notch_max = med + 1.57 * iqr / np.sqrt(N)
return notch_min, notch_max
# output is a list of dicts
bxpstats = []
# convert X to a list of lists
X = _reshape_2D(X, "X")
ncols = len(X)
if labels is None:
labels = itertools.repeat(None)
elif len(labels) != ncols:
raise ValueError(f"The number of labels ({len(labels)}) must match the"
f" number of columns ({ncols}).")
input_whis = whis
for ii, (x, label) in enumerate(zip(X, labels)):
# empty dict
stats = {}
if label is not None:
stats['label'] = label
# restore whis to the input values in case it got changed in the loop
whis = input_whis
# note tricksiness, append up here and then mutate below
bxpstats.append(stats)
# if empty, bail
if len(x) == 0:
stats['fliers'] = np.array([])
stats['mean'] = np.nan
stats['med'] = np.nan
stats['q1'] = np.nan
stats['q3'] = np.nan
stats['iqr'] = np.nan
stats['cilo'] = np.nan
stats['cihi'] = np.nan
stats['whislo'] = np.nan
stats['whishi'] = np.nan
continue
# up-convert to an array, just to be safe
x = np.ma.asarray(x)
x = x.data[~x.mask].ravel()
# arithmetic mean
stats['mean'] = np.mean(x)
# medians and quartiles
q1, med, q3 = np.percentile(x, [25, 50, 75])
# interquartile range
stats['iqr'] = q3 - q1
if stats['iqr'] == 0 and autorange:
whis = (0, 100)
# conf. interval around median
stats['cilo'], stats['cihi'] = _compute_conf_interval(
x, med, stats['iqr'], bootstrap
)
# lowest/highest non-outliers
if np.iterable(whis) and not isinstance(whis, str):
loval, hival = np.percentile(x, whis)
elif np.isreal(whis):
loval = q1 - whis * stats['iqr']
hival = q3 + whis * stats['iqr']
else:
raise ValueError('whis must be a float or list of percentiles')
# get high extreme
wiskhi = x[x = loval]
if len(wisklo) == 0 or np.min(wisklo) > q1:
stats['whislo'] = q1
else:
stats['whislo'] = np.min(wisklo)
# compute a single array of outliers
stats['fliers'] = np.concatenate([
x[x < stats['whislo']],
x[x > stats['whishi']],
])
# add in the remaining stats
stats['q1'], stats['med'], stats['q3'] = q1, med, q3
return bxpstats
#: Maps short codes for line style to their full name used by backends.
ls_mapper = {'-': 'solid', '--': 'dashed', '-.': 'dashdot', ':': 'dotted'}
#: Maps full names for line styles used by backends to their short codes.
ls_mapper_r = {v: k for k, v in ls_mapper.items()}
def contiguous_regions(mask):
"""
Return a list of (ind0, ind1) such that ``mask[ind0:ind1].all()`` is
True and we cover all such regions.
"""
mask = np.asarray(mask, dtype=bool)
if not mask.size:
return []
# Find the indices of region changes, and correct offset
idx, = np.nonzero(mask[:-1] != mask[1:])
idx += 1
# List operations are faster for moderately sized arrays
idx = idx.tolist()
# Add first and/or last index if needed
if mask[0]:
idx = [0] + idx
if mask[-1]:
idx.append(len(mask))
return list(zip(idx[::2], idx[1::2]))
def is_math_text(s):
"""
Return whether the string *s* contains math expressions.
This is done by checking whether *s* contains an even number of
non-escaped dollar signs.
"""
s = str(s)
dollar_count = s.count(r'$') - s.count(r'\$')
even_dollars = (dollar_count > 0 and dollar_count % 2 == 0)
return even_dollars
def _to_unmasked_float_array(x):
"""
Convert a sequence to a float array; if input was a masked array, masked
values are converted to nans.
"""
if hasattr(x, 'mask'):
return np.ma.asanyarray(x, float).filled(np.nan)
else:
return np.asanyarray(x, float)
def _check_1d(x):
"""Convert scalars to 1D arrays; pass-through arrays as is."""
# Unpack in case of e.g. Pandas or xarray object
x = _unpack_to_numpy(x)
# plot requires `shape` and `ndim`. If passed an
# object that doesn't provide them, then force to numpy array.
# Note this will strip unit information.
if (not hasattr(x, 'shape') or
not hasattr(x, 'ndim') or
len(x.shape) < 1):
return np.atleast_1d(x)
else:
return x
def _reshape_2D(X, name):
"""
Use Fortran ordering to convert ndarrays and lists of iterables to lists of
1D arrays.
Lists of iterables are converted by applying `numpy.asanyarray` to each of
their elements. 1D ndarrays are returned in a singleton list containing
them. 2D ndarrays are converted to the list of their *columns*.
*name* is used to generate the error message for invalid inputs.
"""
# Unpack in case of e.g. Pandas or xarray object
X = _unpack_to_numpy(X)
# Iterate over columns for ndarrays.
if isinstance(X, np.ndarray):
X = X.transpose()
if len(X) == 0:
return [[]]
elif X.ndim == 1 and np.ndim(X[0]) == 0:
# 1D array of scalars: directly return it.
return [X]
elif X.ndim in [1, 2]:
# 2D array, or 1D array of iterables: flatten them first.
return [np.reshape(x, -1) for x in X]
else:
raise ValueError(f'{name} must have 2 or fewer dimensions')
# Iterate over list of iterables.
if len(X) == 0:
return [[]]
result = []
is_1d = True
for xi in X:
# check if this is iterable, except for strings which we
# treat as singletons.
if not isinstance(xi, str):
try:
iter(xi)
except TypeError:
pass
else:
is_1d = False
xi = np.asanyarray(xi)
nd = np.ndim(xi)
if nd > 1:
raise ValueError(f'{name} must have 2 or fewer dimensions')
result.append(xi.reshape(-1))
if is_1d:
# 1D array of scalars: directly return it.
return [np.reshape(result, -1)]
else:
# 2D array, or 1D array of iterables: use flattened version.
return result
def violin_stats(X, method=("GaussianKDE", "scott"), points=100, quantiles=None):
"""
Return a list of dictionaries of data which can be used to draw a series
of violin plots.
See the ``Returns`` section below to view the required keys of the
dictionary.
Users can skip this function and pass a user-defined set of dictionaries
with the same keys to `~.axes.Axes.violin` instead of using Matplotlib
to do the calculations. See the *Returns* section below for the keys
that must be present in the dictionaries.
Parameters
----------
X : 1D array or sequence of 1D arrays or 2D array
Sample data that will be used to produce the gaussian kernel density
estimates. Non-finite and masked values are ignored.
Possible values:
- 1D array: Statistics are computed for that array.
- sequence of 1D arrays: Statistics are computed for each array in the sequence.
- 2D array: Statistics are computed for each column in the array.
method : (name, bw_method) or callable,
The method used to calculate the kernel density estimate for each
column of data. Valid values:
- a tuple of the form ``(name, bw_method)`` where *name* currently must
always be ``"GaussianKDE"`` and *bw_method* is the method used to
calculate the estimator bandwidth. Supported values are 'scott',
'silverman' or a float or a callable. If a float, this will be used
directly as `!kde.factor`. If a callable, it should take a
`matplotlib.mlab.GaussianKDE` instance as its only parameter and
return a float.
- a callable with the signature ::
def method(data: ndarray, coords: ndarray) -> ndarray
It should return the KDE of *data* evaluated at *coords*.
.. versionadded:: 3.11
Support for ``(name, bw_method)`` tuple.
points : int, default: 100
Defines the number of points to evaluate each of the gaussian kernel
density estimates at.
quantiles : array-like, default: None
Defines (if not None) a list of floats in interval [0, 1] for each
column of data, which represents the quantiles that will be rendered
for that column of data. Must have 2 or fewer dimensions. 1D array will
be treated as a singleton list containing them.
Returns
-------
list of dict
A list of dictionaries containing the results for each column of data.
The dictionaries contain at least the following:
- coords: A list of scalars containing the coordinates this particular
kernel density estimate was evaluated at.
- vals: A list of scalars containing the values of the kernel density
estimate at each of the coordinates given in *coords*.
- mean: The mean value for this column of data.
- median: The median value for this column of data.
- min: The minimum value for this column of data.
- max: The maximum value for this column of data.
- quantiles: The quantile values for this column of data.
"""
if isinstance(method, tuple):
name, bw_method = method
if name != "GaussianKDE":
raise ValueError(f"Unknown KDE method name {name!r}. The only supported "
'named method is "GaussianKDE"')
def _kde_method(x, coords):
# fallback gracefully if the vector contains only one value
if np.all(x[0] == x):
return (x[0] == coords).astype(float)
kde = mlab.GaussianKDE(x, bw_method)
return kde.evaluate(coords)
method = _kde_method
# List of dictionaries describing each of the violins.
vpstats = []
# Want X to be a list of data sequences
X = _reshape_2D(X, "X")
# Want quantiles to be as the same shape as data sequences
if quantiles is not None and len(quantiles) != 0:
quantiles = _reshape_2D(quantiles, "quantiles")
# Else, mock quantiles if it's none or empty
else:
quantiles = [[]] * len(X)
# quantiles should have the same size as dataset
if len(X) != len(quantiles):
raise ValueError("List of violinplot statistics and quantiles values"
" must have the same length")
# Zip x and quantiles
for (x, quantile) in zip(X, quantiles):
x = np.asarray(x)
x, = delete_masked_points(x)
if len(x) == 0:
vpstats.append({
'vals': np.array([]),
'coords': np.array([]),
'mean': np.nan,
'median': np.nan,
'min': np.nan,
'max': np.nan,
'quantiles': np.array([]),
})
else:
min_val = np.min(x)
max_val = np.max(x)
coords = np.linspace(min_val, max_val, points)
vpstats.append({
'vals': method(x, coords),
'coords': coords,
'mean': np.mean(x),
'median': np.median(x),
'min': min_val,
'max': max_val,
'quantiles': np.atleast_1d(np.percentile(x, 100 * quantile))
})
return vpstats
def pts_to_prestep(x, *args):
"""
Convert continuous line to pre-steps.
Given a set of ``N`` points, convert to ``2N - 1`` points, which when
connected linearly give a step function which changes values at the
beginning of the intervals.
Parameters
----------
x : array
The x location of the steps. May be empty.
y1, ..., yp : array
y arrays to be turned into steps; all must be the same length as ``x``.
Returns
-------
array
The x and y values converted to steps in the same order as the input;
can be unpacked as ``x_out, y1_out, ..., yp_out``. If the input is
length ``N``, each of these arrays will be length ``2N + 1``. For
``N=0``, the length will be 0.
Examples
--------
>>> x_s, y1_s, y2_s = pts_to_prestep(x, y1, y2)
"""
steps = np.zeros((1 + len(args), max(2 * len(x) - 1, 0)))
# In all `pts_to_*step` functions, only assign once using *x* and *args*,
# as converting to an array may be expensive.
steps[0, 0::2] = x
steps[0, 1::2] = steps[0, 0:-2:2]
steps[1:, 0::2] = args
steps[1:, 1::2] = steps[1:, 2::2]
return steps
def pts_to_poststep(x, *args):
"""
Convert continuous line to post-steps.
Given a set of ``N`` points convert to ``2N + 1`` points, which when
connected linearly give a step function which changes values at the end of
the intervals.
Parameters
----------
x : array
The x location of the steps. May be empty.
y1, ..., yp : array
y arrays to be turned into steps; all must be the same length as ``x``.
Returns
-------
array
The x and y values converted to steps in the same order as the input;
can be unpacked as ``x_out, y1_out, ..., yp_out``. If the input is
length ``N``, each of these arrays will be length ``2N + 1``. For
``N=0``, the length will be 0.
Examples
--------
>>> x_s, y1_s, y2_s = pts_to_poststep(x, y1, y2)
"""
steps = np.zeros((1 + len(args), max(2 * len(x) - 1, 0)))
steps[0, 0::2] = x
steps[0, 1::2] = steps[0, 2::2]
steps[1:, 0::2] = args
steps[1:, 1::2] = steps[1:, 0:-2:2]
return steps
def pts_to_midstep(x, *args):
"""
Convert continuous line to mid-steps.
Given a set of ``N`` points convert to ``2N`` points which when connected
linearly give a step function which changes values at the middle of the
intervals.
Parameters
----------
x : array
The x location of the steps. May be empty.
y1, ..., yp : array
y arrays to be turned into steps; all must be the same length as
``x``.
Returns
-------
array
The x and y values converted to steps in the same order as the input;
can be unpacked as ``x_out, y1_out, ..., yp_out``. If the input is
length ``N``, each of these arrays will be length ``2N``.
Examples
--------
>>> x_s, y1_s, y2_s = pts_to_midstep(x, y1, y2)
"""
steps = np.zeros((1 + len(args), 2 * len(x)))
x = np.asanyarray(x)
steps[0, 1:-1:2] = steps[0, 2::2] = (x[:-1] + x[1:]) / 2
steps[0, :1] = x[:1] # Also works for zero-sized input.
steps[0, -1:] = x[-1:]
steps[1:, 0::2] = args
steps[1:, 1::2] = steps[1:, 0::2]
return steps
STEP_LOOKUP_MAP = {'default': lambda x, y: (x, y),
'steps': pts_to_prestep,
'steps-pre': pts_to_prestep,
'steps-post': pts_to_poststep,
'steps-mid': pts_to_midstep}
def index_of(y):
"""
A helper function to create reasonable x values for the given *y*.
This is used for plotting (x, y) if x values are not explicitly given.
First try ``y.index`` (assuming *y* is a `pandas.Series`), if that
fails, use ``range(len(y))``.
This will be extended in the future to deal with more types of
labeled data.
Parameters
----------
y : float or array-like
Returns
-------
x, y : ndarray
The x and y values to plot.
"""
try:
return y.index.to_numpy(), y.to_numpy()
except AttributeError:
pass
try:
y = _check_1d(y)
except (VisibleDeprecationWarning, ValueError):
# NumPy 1.19 will warn on ragged input, and we can't actually use it.
pass
else:
return np.arange(y.shape[0], dtype=float), y
raise ValueError('Input could not be cast to an at-least-1D NumPy array')
def safe_first_element(obj):
"""
Return the first element in *obj*.
This is a type-independent way of obtaining the first element,
supporting both index access and the iterator protocol.
"""
if isinstance(obj, collections.abc.Iterator):
# needed to accept `array.flat` as input.
# np.flatiter reports as an instance of collections.Iterator but can still be
# indexed via []. This has the side effect of re-setting the iterator, but
# that is acceptable.
try:
return obj[0]
except TypeError:
pass
raise RuntimeError("matplotlib does not support generators as input")
return next(iter(obj))
def _safe_first_finite(obj):
"""
Return the first finite element in *obj* if one is available and skip_nonfinite is
True. Otherwise, return the first element.
This is a method for internal use.
This is a type-independent way of obtaining the first finite element, supporting
both index access and the iterator protocol.
"""
def safe_isfinite(val):
if val is None:
return False
try:
return math.isfinite(val)
except (TypeError, ValueError):
# if the outer object is 2d, then val is a 1d array, and
# - math.isfinite(numpy.zeros(3)) raises TypeError
# - math.isfinite(torch.zeros(3)) raises ValueError
pass
try:
return np.isfinite(val) if np.isscalar(val) else True
except TypeError:
# This is something that NumPy cannot make heads or tails of,
# assume "finite"
return True
if isinstance(obj, np.flatiter):
# TODO do the finite filtering on this
return obj[0]
elif isinstance(obj, collections.abc.Iterator):
raise RuntimeError("matplotlib does not support generators as input")
else:
for val in obj:
if safe_isfinite(val):
return val
return safe_first_element(obj)
def sanitize_sequence(data):
"""
Convert dictview objects to list. Other inputs are returned unchanged.
"""
return (list(data) if isinstance(data, collections.abc.MappingView)
else data)
def _resize_sequence(seq, N):
"""
Trim the given sequence to exactly N elements.
If there are more elements in the sequence, cut it.
If there are less elements in the sequence, repeat them.
Implementation detail: We maintain type stability for the output for
N len(seq); this was good
enough for the present use cases but is not a fixed design decision.
"""
num_elements = len(seq)
if N == num_elements:
return seq
elif N < num_elements:
return seq[:N]
else:
return list(itertools.islice(itertools.cycle(seq), N))
def normalize_kwargs(kw, alias_mapping=None):
"""
Helper function to normalize kwarg inputs.
Parameters
----------
kw : dict or None
A dict of keyword arguments. None is explicitly supported and treated
as an empty dict, to support functions with an optional parameter of
the form ``props=None``.
alias_mapping : Artist subclass or Artist instance
A mapping between a canonical name to a list of aliases, in order of
precedence from lowest to highest.
If the canonical value is not in the list it is assumed to have the
highest priority.
If an Artist subclass or instance is passed, use its properties alias
mapping.
Raises
------
TypeError
To match what Python raises if invalid arguments/keyword arguments are
passed to a callable.
"""
from matplotlib.artist import Artist
# deal with default value of alias_mapping
if (isinstance(alias_mapping, type) and issubclass(alias_mapping, Artist)
or isinstance(alias_mapping, Artist)):
alias_to_prop = getattr(alias_mapping, "_alias_to_prop", {})
else:
if alias_mapping is None:
alias_mapping = {}
_api.warn_deprecated("3.11", message=(
"Passing a dict or None as alias_mapping to normalize_kwargs is "
"deprecated since %(since)s and support will be removed "
"%(removal)s; pass an Artist instance or type instead."))
# Convert old format to new format.
alias_to_prop = {alias: prop for prop, aliases in alias_mapping.items()
for alias in aliases}
if kw is None:
return {}
canonicalized = {alias_to_prop.get(k, k): v for k, v in kw.items()}
if len(canonicalized) == len(kw):
return canonicalized
canonical_to_seen = {}
for k in kw:
canonical = alias_to_prop.get(k, k)
if canonical in canonical_to_seen:
raise TypeError(f"Got both {canonical_to_seen[canonical]!r} and "
f"{k!r}, which are aliases of one another")
canonical_to_seen[canonical] = k
@contextlib.contextmanager
def _lock_path(path):
"""
Context manager for locking a path.
Usage::
with _lock_path(path):
...
Another thread or process that attempts to lock the same path will wait
until this context manager is exited.
The lock is implemented by creating a temporary file in the parent
directory, so that directory must exist and be writable.
"""
path = Path(path)
lock_path = path.with_name(path.name + ".matplotlib-lock")
retries = 50
sleeptime = 0.1
for _ in range(retries):
try:
with lock_path.open("xb"):
break
except FileExistsError:
time.sleep(sleeptime)
else:
raise TimeoutError("""\
Lock error: Matplotlib failed to acquire the following lock file:
{}
This maybe due to another process holding this lock file. If you are sure no
other Matplotlib process is running, remove this file and try again.""".format(
lock_path))
try:
yield
finally:
lock_path.unlink()
def _topmost_artist(
artists,
_cached_max=functools.partial(max, key=operator.attrgetter("zorder"))):
"""
Get the topmost artist of a list.
In case of a tie, return the *last* of the tied artists, as it will be
drawn on top of the others. `max` returns the first maximum in case of
ties, so we need to iterate over the list in reverse order.
"""
return _cached_max(reversed(artists))
def _str_equal(obj, s):
"""
Return whether *obj* is a string equal to string *s*.
This helper solely exists to handle the case where *obj* is a numpy array,
because in such cases, a naive ``obj == s`` would yield an array, which
cannot be used in a boolean context.
"""
return isinstance(obj, str) and obj == s
def _str_lower_equal(obj, s):
"""
Return whether *obj* is a string equal, when lowercased, to string *s*.
This helper solely exists to handle the case where *obj* is a numpy array,
because in such cases, a naive ``obj == s`` would yield an array, which
cannot be used in a boolean context.
"""
return isinstance(obj, str) and obj.lower() == s
def _array_perimeter(arr):
"""
Get the elements on the perimeter of *arr*.
Parameters
----------
arr : ndarray, shape (M, N)
The input array.
Returns
-------
ndarray, shape (2*(M - 1) + 2*(N - 1),)
The elements on the perimeter of the array::
[arr[0, 0], ..., arr[0, -1], ..., arr[-1, -1], ..., arr[-1, 0], ...]
Examples
--------
>>> i, j = np.ogrid[:3, :4]
>>> a = i*10 + j
>>> a
array([[ 0, 1, 2, 3],
[10, 11, 12, 13],
[20, 21, 22, 23]])
>>> _array_perimeter(a)
array([ 0, 1, 2, 3, 13, 23, 22, 21, 20, 10])
"""
# note we use Python's half-open ranges to avoid repeating
# the corners
forward = np.s_[0:-1] # [0 ... -1)
backward = np.s_[-1:0:-1] # [-1 ... 0)
return np.concatenate((
arr[0, forward],
arr[forward, -1],
arr[-1, backward],
arr[backward, 0],
))
def _unfold(arr, axis, size, step):
"""
Append an extra dimension containing sliding windows along *axis*.
All windows are of size *size* and begin with every *step* elements.
Parameters
----------
arr : ndarray, shape (N_1, ..., N_k)
The input array
axis : int
Axis along which the windows are extracted
size : int
Size of the windows
step : int
Stride between first elements of subsequent windows.
Returns
-------
ndarray, shape (N_1, ..., 1 + (N_axis-size)/step, ..., N_k, size)
Examples
--------
>>> i, j = np.ogrid[:3, :7]
>>> a = i*10 + j
>>> a
array([[ 0, 1, 2, 3, 4, 5, 6],
[10, 11, 12, 13, 14, 15, 16],
[20, 21, 22, 23, 24, 25, 26]])
>>> _unfold(a, axis=1, size=3, step=2)
array([[[ 0, 1, 2],
[ 2, 3, 4],
[ 4, 5, 6]],
[[10, 11, 12],
[12, 13, 14],
[14, 15, 16]],
[[20, 21, 22],
[22, 23, 24],
[24, 25, 26]]])
"""
new_shape = [*arr.shape, size]
new_strides = [*arr.strides, arr.strides[axis]]
new_shape[axis] = (new_shape[axis] - size) // step + 1
new_strides[axis] = new_strides[axis] * step
return np.lib.stride_tricks.as_strided(arr,
shape=new_shape,
strides=new_strides,
writeable=False)
def _array_patch_perimeters(x, rstride, cstride):
"""
Extract perimeters of patches from *arr*.
Extracted patches are of size (*rstride* + 1) x (*cstride* + 1) and
share perimeters with their neighbors. The ordering of the vertices matches
that returned by ``_array_perimeter``.
Parameters
----------
x : ndarray, shape (N, M)
Input array
rstride : int
Vertical (row) stride between corresponding elements of each patch
cstride : int
Horizontal (column) stride between corresponding elements of each patch
Returns
-------
ndarray, shape (N/rstride * M/cstride, 2 * (rstride + cstride))
"""
assert rstride > 0 and cstride > 0
assert (x.shape[0] - 1) % rstride == 0
assert (x.shape[1] - 1) % cstride == 0
# We build up each perimeter from four half-open intervals. Here is an
# illustrated explanation for rstride == cstride == 3
#
# T T T R
# L R
# L R
# L B B B
#
# where T means that this element will be in the top array, R for right,
# B for bottom and L for left. Each of the arrays below has a shape of:
#
# (number of perimeters that can be extracted vertically,
# number of perimeters that can be extracted horizontally,
# cstride for top and bottom and rstride for left and right)
#
# Note that _unfold doesn't incur any memory copies, so the only costly
# operation here is the np.concatenate.
top = _unfold(x[:-1:rstride, :-1], 1, cstride, cstride)
bottom = _unfold(x[rstride::rstride, 1:], 1, cstride, cstride)[..., ::-1]
right = _unfold(x[:-1, cstride::cstride], 0, rstride, rstride)
left = _unfold(x[1:, :-1:cstride], 0, rstride, rstride)[..., ::-1]
return (np.concatenate((top, right, bottom, left), axis=2)
.reshape(-1, 2 * (rstride + cstride)))
@contextlib.contextmanager
def _setattr_cm(obj, **kwargs):
"""
Temporarily set some attributes; restore original state at context exit.
"""
sentinel = object()
origs = {}
for attr in kwargs:
orig = getattr(obj, attr, sentinel)
if attr in obj.__dict__ or orig is sentinel:
# if we are pulling from the instance dict or the object
# does not have this attribute we can trust the above
origs[attr] = orig
else:
# if the attribute is not in the instance dict it must be
# from the class level
cls_orig = getattr(type(obj), attr)
# if we are dealing with a property (but not a general descriptor)
# we want to set the original value back.
if isinstance(cls_orig, property):
origs[attr] = orig
# otherwise this is _something_ we are going to shadow at
# the instance dict level from higher up in the MRO. We
# are going to assume we can delattr(obj, attr) to clean
# up after ourselves. It is possible that this code will
# fail if used with a non-property custom descriptor which
# implements __set__ (and __delete__ does not act like a
# stack). However, this is an internal tool and we do not
# currently have any custom descriptors.
else:
origs[attr] = sentinel
try:
for attr, val in kwargs.items():
setattr(obj, attr, val)
yield
finally:
for attr, orig in origs.items():
if orig is sentinel:
delattr(obj, attr)
else:
setattr(obj, attr, orig)
class _OrderedSet(collections.abc.MutableSet):
def __init__(self):
self._od = collections.OrderedDict()
def __contains__(self, key):
return key in self._od
def __iter__(self):
return iter(self._od)
def __len__(self):
return len(self._od)
def add(self, key):
self._od.pop(key, None)
self._od[key] = None
def discard(self, key):
self._od.pop(key, None)
# Agg's buffers are unmultiplied RGBA8888, which neither PyQt